level: research
in manufacturing, worker skills are critical. certifications can expire if not used, and training takes time away from production. this creates a trade-off: train workers now to meet future needs, or use them for current output. a new controller addresses this by solving a mixed-integer program at each shift, deciding production, inventory, backlog, and training actions. it uses a terminal value that prices gaps in certified capacity at the planning horizon, making decisions interpretable.
the controller was tested on synthetic scenarios called skillchain-gym. these included sudden new skill requirements, demand spikes, worker absences, and varying forecast quality. the method replans after each period, applying only the first action. results show it can maintain throughput and reduce backlog compared to simpler rules, even when shocks are unannounced. the approach explicitly models binary certification status and hard eligibility constraints, ensuring only qualified workers are assigned to tasks.
the work highlights how integrating skill dynamics into production planning can make supply chains more resilient. by treating training as a control variable, the system adapts to changing conditions without manual intervention. the interpretable terminal value helps managers understand the cost of skill gaps. this bridges operations research and workforce management, offering a practical tool for industries with complex skill requirements.
why it matters: it shows how ai-driven planning can jointly optimize production and workforce training, reducing disruptions in skill-constrained environments.